app.py

import streamlit as st 
from transformers import pipeline
import pandas as pd
import numpy as np
from transformers import BertTokenizerFast

st.set_page_config(page_title="AI Project", page_icon=":tada:", layout="wide")

pre_trained_models = ["finiteautomata/bertweet-base-sentiment-analysis",
                      'bhadresh-savani/distilbert-base-uncased-emotion',
                      "nlptown/bert-base-multilingual-uncased-sentiment"]

fine_tuned_model = "corykhal/twitter-finetuned"

fine_tuned_labels = {"LABEL_0": "Toxic",
                     "LABEL_1": "Severe Toxic",
                     "LABEL_2": "Obscene",
                     "LABEL_3": "Threat",
                     "LABEL_4": "Insult",
                     "LABEL_5": "Identity Hate"}

models = pre_trained_models + [fine_tuned_model]

with st.container():
    st.title("Hello! Welcome to the Sentiment Analysis App :wave:")
    st.header("By: Cory Khalilollahi")

with st.container():
    st.write("---")

    text = st.text_input("Please enter the text of a tweet to use for the sentiment analysis:",
                         value="Hello! It is a pleasure to meet you!")

    st.write("---")

    st.write("The first 3 models in the select box are for any sentiment analysis.")
    st.write("The last model (which is finetuned) in the select box is specifically for toxicity analysis.")

    model = st.selectbox("Please select one of the following pre-trained models:", models)

    

with st.container():
    st.write("---")
    if st.button("Analyze!"):
        # If the user selects a pre-trained model
        if model in pre_trained_models:
            analysis = pipeline("sentiment-analysis", model=model)
            result = analysis(text)
            sentiment = result[0]["label"] 
            score = result[0]["score"]
            
            # Create a table for the label and score
            data = pd.DataFrame({"Tweet": [text], "Sentiment": [sentiment], "Confidence Score": [score]})

            st.table(data)
        
        # If the user selects the finetuned model
        else:
            analysis = pipeline("sentiment-analysis", model=model, top_k=2)
            result = analysis(text)
            
            # Get the top two labels and scores
            sentiment1 = fine_tuned_labels[result[0][0]["label"]]
            score1 = result[0][0]["score"]

            sentiment2 = fine_tuned_labels[result[0][1]["label"]]
            score2 = result[0][1]["score"]

            # Create a table for the labels and scores
            data = pd.DataFrame({"Tweet": [text], "Toxic": [sentiment1], "Toxic Confidence Score": [score1],
                                 "Type of Toxicity": [sentiment2], "Toxicity Type Confidence Score": [score2]})


            st.table(data)